ETH Zurich engineers develop autonomous mobile robotic hand

Engineers at ETH Zurich’s Soft Robotics Lab have developed a detached, mobile robotic hand capable of independent movement and balance. Unlike previous designs that required symmetrical fingers for locomotion, this research successfully trained a commercial WUJI robotic hand—which features unequal finger lengths and an opposing thumb—to crawl and navigate autonomously.

The 818-gram device is a self-contained unit, equipped with a Raspberry Pi, motion sensors, and a lithium-polymer battery mounted to its palm. To overcome the stability challenges posed by its irregular shape, the team used deep reinforcement learning within NVIDIA’s Isaac lab simulation platform. This training allowed the hand to determine how to lift individual digits to maintain forward progress.

In testing, the hand demonstrated versatility by crawling across 14 different indoor and outdoor surfaces. It also features a recovery algorithm that enables the robot to right itself back onto its fingertips in 84% of hardware fall attempts. Researchers suggest these mobile robotic hands could be highly effective in confined or difficult-to-access operational environments.